Children, Social Media and AI

Australia banned under-16s from ten platforms on 10 December 2025. Six months of published evidence shows the ban has not moved usage, and the cost of the attempt is a generation held back from the tools it will work with. Sovereign Australia’s position: replace the age ban with one national parental control system, built with the platforms rather than forced on them, and put AI into classrooms rather than out of reach.

1. What the law does

The Online Safety Amendment (Social Media Minimum Age) Act 2024 amended the Online Safety Act 2021 to require age-restricted platforms to take reasonable steps to prevent Australians under 16 from holding accounts. It passed both houses on 29 November 2024 — the Senate vote was 34 to 19 — and commenced on 10 December 2025, the first law of its kind in the world.

Ten platforms are in scope: Facebook, Instagram, Kick, Reddit, Snapchat, Threads, TikTok, Twitch, X and YouTube. Others assessed as lower risk are not, including WhatsApp, Discord, Roblox and Pinterest. Penalties for systemic failure reach $49.5 million. The eSafety Commissioner writes the guidelines on what reasonable steps means, and issued compulsory information notices to ten platforms on 11 December 2025. An independent review of the Act’s operation is required within two years.

The Act contains no parental consent provision. A parent who judges their fifteen-year-old ready for a supervised account has no mechanism to allow it, and a ten-year-old and a fifteen-year-old are treated identically. Two High Court challenges are on foot; Reddit filed on 12 December 2025, arguing the law burdens the implied freedom of political communication because those excluded will be voters within a few years.

2. What has happened since

Three separate measurements of the first six months are on the public record, and they agree.

The eSafety Commissioner’s own compliance figures in March 2026 found roughly seven in ten under-16s still held accounts on Facebook, Instagram, Snapchat and TikTok.

A peer-reviewed study in the British Medical Journal, published June 2026, surveyed 408 Australians aged 12 to 17 immediately before the restrictions and again three months after. More than 85 per cent of under-16s were still using platforms covered by the Act at follow-up, most of them on their own accounts. Fake accounts accounted for 15 to 19 per cent and private browsing for 6 to 11 per cent. Daily use was unchanged among 12 and 13 year olds, fell from 78 to 69 per cent among 14 and 15 year olds, and rose from 80 to 89 per cent among those 16 and over. The authors found insufficient evidence of any early substantial effect, and described the period as one of limited implementation, incomplete compliance and substantial circumvention.

An April 2026 working paper by a team including Cass Sunstein and Angela Duckworth, titled Why Bans Fail, put compliance among 14 and 15 year olds at about 27 per cent. Three-quarters of the teenagers surveyed considered circumvention easy. Nearly two-thirds had never had an account removed.

The Government’s published measure of success is different: 4.7 million accounts revoked by January 2026. That is a count of deactivations, not a count of children off the platforms, and the three measurements above are what happened to the second number. In March 2026 the Communications Minister said the laws were not failing but that the platforms were failing to obey them.

On 28 June 2026 the Government announced what strengthening means. The maximum penalty for breaches of the minimum-age law is to be doubled to $99 million — the legislated maximum was 150,000 penalty units, which the eSafety Commissioner currently states as $54.6 million — explicitly to bring it into line with competition and consumer law. The eSafety Commissioner’s information-gathering powers are to be expanded, including the power to compel documents and evidence from platforms about how they enforce the ban, and to require information from third-party age-verification and digital-identity providers. The Minister said platforms were doing the bare minimum to get by and that the Government would not back down. The Prime Minister said there were still too many children on social media. The Government’s count at that point was more than five million under-16 accounts removed, deactivated or restricted.

3. Why prohibition of a digital product does not hold

The mechanism of failure is not complicated and it was predictable. A prohibited physical good has to cross a border or a counter. A social media account has to cross nothing: it needs a birth year typed into a form, a parent’s login, an older sibling’s account, a second phone, or a browser window. Enforcement lands on the platform, the child is the party being kept out, and the child has both the motive and the means to get in. That is the same structure that has defeated every prohibition regime applied to something people want and can obtain privately.

There is a second, harder problem. Keeping under-16s out requires knowing who is 16, which requires checking the age of every user, including every adult. That is why the tighter the enforcement, the larger the identity infrastructure — and the better the privacy protections, the weaker the enforcement. Strengthening the Act in response to poor compliance necessarily means more age verification of the whole population, not just of children. Australians were not asked to accept that trade, and it does not become smaller by being called an online safety measure.

Circumvention is not neutral. It makes the child less safe than before the ban. A platform applies its youth protections to accounts it knows belong to a minor: restricted messaging, limited recommendations, no adult content, advertising limits, stricter defaults. A child who gets back in with a false birth year holds an adult account, and every one of those protections is switched off, because the platform has been told it is dealing with an adult. The measured circumvention rate is therefore not a compliance statistic. It is a count of children who moved from a supervised teen account to an unsupervised adult one, with adult content, adult contact and adult targeting attached. The BMJ study put fake-account use at 15 to 19 per cent within three months. The law was passed to reduce exposure; circumvention produces more of it.

The precedent is alcohol prohibition, and it is the closest one available. The United States prohibited alcohol from 1920 under the Volstead Act. Consumption fell sharply at first and then recovered substantially; supply moved to criminal production; and because no one was regulating the product, people were poisoned by what they drank. After thirteen years the answer was not tighter prohibition. It was repeal, in 1933, and replacement with a regulated market: a minimum purchase age, licensed sellers, product standards, taxation, and enforcement directed at the seller rather than the drinker.

That is the model Australia now applies to alcohol, tobacco, gambling and driving, and it works well enough that nobody proposes returning to prohibition for any of them. A fourteen-year-old cannot buy spirits at a bottle shop, and the country is not attempting to make spirits unobtainable. Nothing about a digital product makes it the exception — except that its controls are cheaper, faster and far more precise than anything a bottle shop can do.

Third, exclusion defers a problem rather than solving it. A child kept off a platform until the day they turn 16 arrives there with no supervised experience, no accumulated judgement and no adult in the room, at the age when their account is fully unrestricted. The alternative is a graduated introduction under controls that tighten and loosen with age, which is how every other risk presented to children is handled — roads, water, tools, money and vehicles.

4. What exclusion costs

The platforms in scope are not only entertainment. They are where a fifteen-year-old learns to code from a YouTube channel, follows a trade apprentice on Instagram, runs a small business page, joins a Reddit forum on their subject, or watches a scientist explain something no local school teaches. Removing all of it removes the harm and the education together, because the Act draws no distinction between them.

Three specific costs follow. A young person operating commercially or in advocacy loses working tools with no individualised assessment. Digital and media literacy — the ability to identify a scam, a manipulated image, a bot, a bad-faith argument — is learned by supervised use, and cannot be taught to a cohort excluded from the medium. And the country’s children spend the years when they learn fastest formally locked out of the dominant communication technology of their working lives, in a labour market where every other country’s children are not.

None of that argues the harms are imaginary. Bullying, contact by adults, algorithmic amplification of self-harm and eating-disorder material, and engagement-optimised design aimed at children are real and documented. They are arguments for controlling the design of the product, which is the section that follows.

5. One national parental control system

The controls already exist and are already commercial. YouTube Kids operates a curated catalogue, no general search, timers and parental dashboards. Console and phone operating systems ship with screen-time limits, app allow-lists, purchase locks and content ratings. Schools run filtered networks. Nothing here needs inventing.

What does not exist is one place for a parent to use it. The settings are scattered across ten platforms and two operating systems, each with different names, different defaults and different menus, and a parent has to find and maintain all of them separately. That is the actual gap, and it is an engineering problem rather than a legislative one.

Sovereign Australia’s position: the platforms are required to build one national system through which a parent controls their own child’s access — one standard, one enrolment, one dashboard, working the same way across every platform in Australia. Through it a parent sets:

• Who the account belongs to. A child’s account is enrolled as a child’s account and stays that way across platforms, so a false birth year on one service does not undo the settings on the others.
• Time. Daily limits, quiet hours and school-hours restrictions, set once and applied everywhere, enforced in the product and not defeated by reinstalling the app.
• Content, by age band. The strictest tier a curated catalogue on the YouTube Kids model, loosening in defined steps as the child gets older, with the parent choosing the step.
• Contact. No unsolicited adult contact, no discoverability of a child by strangers, direct messaging off unless the parent turns it on.
• Design. For an enrolled child account: no engagement-optimised recommender feed, no infinite scroll, no autoplay, no streak or notification mechanics built to pull a child back, and no targeted advertising or data profiling.
• Age, once. Signalled at the device or operating-system level and attested to platforms as a yes-or-no, so no platform holds identity documents and no central identity register is built.

Built with the companies, not against them

This is the part the current law gets backwards. The Act makes a platform liable for a child’s deception — the company is fined because a fourteen-year-old typed a false birth year — and the enforcement language treats the industry as an offender to be brought to heel. The predictable result is a legal fight rather than a build: two High Court challenges, compulsory information notices, ministerial statements about companies failing to obey, and engineering effort spent on compliance defence instead of on tools parents can use.

The companies in question have the largest client-side engineering capability on earth, and they already ship parental controls in other jurisdictions. Sovereign Australia’s approach is to convene them, specify the outcome the system has to deliver, agree the technical standard with them, and set a delivery date. The regulator’s job is to certify that the system works and to publish who has connected to it. Penalty exposure attaches to refusing to build or connect — not to whether a determined child got around it.

The distinction matters practically. A company being sued has lawyers on the problem; a company that co-authored the standard has engineers on it. One national system also costs the industry less than ten separate compliance regimes and an unbounded liability for teenage dishonesty, which is why it is achievable rather than merely desirable.

The Government has chosen the opposite course, and Sovereign Australia’s position is that it is the wrong one. Six months of evidence showed the law was not keeping children off the platforms. The response, announced on 28 June 2026, was to double the maximum penalty to $99 million and expand the regulator’s power to compel documents from platforms and from age-verification and digital-identity providers. Nothing in that package builds a control a parent can use. It raises the price of a failure the platforms cannot fully prevent, because the conduct being punished is a child entering a false birth date.

The language went with it: platforms accused of tricks and of doing the bare minimum, ministers declaring they will not back down, companies described as targets to be held to account. One of the affected companies had already called the legislation rushed and said it misunderstood its platform. That is the vocabulary of prosecution, not of procurement, and it produces what prosecution produces — litigation, minimum compliance, and no shared system at the end of it.

Doubling a fine is not a policy instrument for this problem. It is a statement about resolve. Sovereign Australia would spend the same political capital convening the companies, specifying the system, and holding them to a delivery date — because the measure that matters to a parent is whether the controls exist and work, not how large a penalty was announced.

The criminal law is unchanged and enforced: grooming, child abuse material, sexual extortion and non-consensual deepfakes are crimes and are prosecuted as crimes, with the investigative capacity funded to do it. That is where the coercive power belongs — on the adults committing offences against children, not on the companies for failing to guess a birthday.

6. AI: utilised, not banned

The same instinct that produced the age ban is being applied to AI in classrooms, and the same result follows. A student barred from AI at school uses it at home, unsupervised, without instruction and without disclosure. A student taught to use it learns what it is good at, where it invents, how to check it and how to say when they used it.

Australia already has the instrument. Education Ministers approved the Australian Framework for Generative AI in Schools on 5 October 2023 — six principles and 25 guiding statements across teaching and learning, wellbeing, transparency, fairness, accountability, and privacy and security — implemented from Term 1 2024, with the 2024 review endorsed by Ministers in June 2025. It is two pages of guidance, and it is not a rollout.

Delivery is consequently uneven and depends on the state a child lives in. South Australia has taken EdChat to all public high schools. Queensland is rolling out Corella statewide. Western Australia has trialled AI for teacher workload reduction in eight schools. Victoria requires parental opt-in for tools handling personal information and bans AI-generated depictions of students and staff. Tasmania permits approved tools only, the Northern Territory works to risk-management guidelines, and the ACT is introducing tools in stages. Some children are learning with these tools daily; others are not seeing them.

Sovereign Australia’s position: a funded national rollout, not a framework. AI literacy taught in the curriculum from primary school, at the same level of seriousness as reading a graph or checking a source. Teacher professional development paid for and timetabled, with AI used first on the administrative load — planning, drafting, marking preparation, reporting — because that is where the hours come back to teaching. Assessment redesigned so that what is assessed is what a student can do with the tools and without them, rather than pretending the tools are absent. Approved-tool catalogues with published privacy terms, no sale of student data, no AI-generated depictions of students or staff, disclosure when AI is used by staff or students, and high-stakes marking done by humans.

7. What would have to change

• Repeal the minimum-age provision of the Online Safety Act and replace it with the national parental control system at section 5.
• Legislate the obligation to build and connect, not the obligation to police children: platforms operating in Australia must implement the agreed standard by a set date, and penalty exposure attaches to refusing to build or connect.
• Convene the industry to co-author the technical standard — a published outcome specification, agreed with the platforms and the operating-system vendors, with the regulator certifying conformance and publishing who has connected.
• Restore parental consent and supervision as the mechanism, not an exception to a ban.
• Build age assurance at the device level as a private attestation, with a statutory prohibition on retaining identity documents or building a central age or identity register.
• Fund the national AI-in-schools rollout: approved tools, teacher professional development, curriculum-embedded AI literacy from primary school, and assessment redesign.
• Publish the evidence quarterly. Usage by age band, compliance rates, harm reports and time-on-platform, so the policy is measured on what children actually experience rather than on accounts deactivated.

8. Where the parties stand

Party positions are taken from the recorded votes on the Online Safety Amendment (Social Media Minimum Age) Bill 2024 (House 27 November 2024; Senate 34–19 on 28 November 2024, with the Greens, One Nation’s Malcolm Roberts, and several crossbenchers opposed, and Liberal and National senators crossing the floor), and from party statements to July 2026. “No stated position” means none was found on the public record. This table states each party’s position, not this party’s characterisation of it.

9. Summary

Australia legislated the world’s first social media age ban, and six months of independent measurement found most under-16s still on the platforms, reached through older accounts, fake accounts and private browsers. The published response, on 28 June 2026, was to double the maximum penalty to $99 million and expand the regulator’s powers — which raises the price of a failure the platforms cannot fully prevent, and still means checking the age of every adult to keep out children who are getting in anyway — and the children who get in do so on adult accounts, with every youth protection switched off. Alcohol prohibition ended the same way, and was replaced with a minimum age, licensed sellers, product standards and enforcement aimed at the seller.

The controls that would work already exist commercially: curated content tiers, time limits, contact restrictions and no engagement-optimised feed for a child. What is missing is one place for a parent to use them. Sovereign Australia requires the platforms to build a single national system — one standard, one enrolment, one dashboard — through which a parent sets what their own child can reach, for how long and at what hours, applied the same way across every platform. Age is assured once at the device, not by handing documents to ten companies. And the system is specified and built with the industry rather than extracted from it by prosecution: a company being sued puts lawyers on the problem, a company that co-authored the standard puts engineers on it. AI is taught in classrooms, with the teachers funded and trained to teach it, rather than left to children to use unsupervised at home.

Australian children will work their whole lives in an economy built on these tools. The state’s job is to control how the tools behave toward a child, and to teach the child to use them. It is not to make them unlawful and call that protection.

10. Sources

1. Online Safety Amendment (Social Media Minimum Age) Act 2024 (Cth), amending the Online Safety Act 2021; parliamentary records of the Bill (introduced 21 November 2024; House third reading 27 November 2024; Senate 34 votes to 19 on 28 November 2024; passed both houses 29 November 2024) — the obligation on age-restricted platforms to take reasonable steps to prevent under-16 account holders, the absence of a parental consent pathway, the eSafety Commissioner’s role in setting the reasonable-steps guidelines, and penalties to $49.5 million for systemic failure. Cited in sections 1, 7 and 8.
2. eSafety Commissioner determinations and enforcement material, December 2025 to March 2026 — commencement on 10 December 2025; the ten platforms in scope (Facebook, Instagram, Kick, Reddit, Snapchat, Threads, TikTok, Twitch, X, YouTube) and services assessed as lower risk and excluded (including WhatsApp, Discord, Roblox, Pinterest); compulsory information notices issued to ten platforms on 11 December 2025; and the March 2026 compliance figures indicating approximately seven in ten under-16s still holding accounts on Facebook, Instagram, Snapchat and TikTok. Cited in sections 1 and 2.
3. Assessing the early effects of Australia’s Social Media Minimum Age Act on adolescent social media use, The British Medical Journal, June 2026 — a survey of 408 Australians aged 12 to 17 before commencement and at three months; more than 85 per cent of under-16s still using platforms covered by the Act at follow-up, mostly on their own accounts; fake-account use 15 to 19 per cent and private browsing 6 to 11 per cent; daily use stable among 12 to 13 year olds, 78 to 69 per cent among 14 to 15 year olds, and 80 to 89 per cent among those 16 and over; and the authors’ conclusion of insufficient evidence of early substantial effects, in a period of limited implementation, incomplete compliance and substantial circumvention. Cited in sections 2 and 9.
4. Why Bans Fail, working paper, April 2026, by a team including Cass Sunstein and Angela Duckworth — compliance among 14 and 15 year olds estimated at about 27 per cent; approximately three-quarters of teenagers surveyed considering circumvention easy; and nearly two-thirds never having had an account removed. Working paper, not peer-reviewed at the time of writing. Cited in sections 2 and 9.
5. Australian Government statements and contemporary reporting, January to June 2026 — approximately 4.7 million accounts revoked as reported by the Minister for Communications in January 2026; the government-commissioned finding that 96 per cent of Australian children aged 10 to 15 used social media before commencement; the Minister’s March 2026 statement that the laws were not failing but that platforms were failing to obey them; and the Prime Minister’s statement to Parliament in late June 2026 that the Government was working as a priority to strengthen the law. Cited in sections 2 and 9.
6. High Court proceedings commenced from 12 December 2025, including the application filed by Reddit — the contention that the minimum-age regime burdens the implied constitutional freedom of political communication, given that those excluded reach voting age within a few years; and the requirement in the Act for an independent review of its operation within two years. Cited in section 1.
7. Analysis of the age-assurance trade-off, including commentary by Michael Geist, July 2026 — the observation that enforcing a minimum age requires age checking of the whole user population, such that stronger privacy protection produces weaker enforcement and stronger enforcement requires broader identity verification. Cited in section 3.
8. Commentary on the design of the regime, including the Bloomsbury Intelligence and Security Institute assessment, January 2026 — the removal of parental consent and family discretion; the application of identical restrictions to a ten-year-old and a fifteen-year-old; the loss of working tools by young people engaged in business, journalism or advocacy without individualised assessment; and the effect of producing sudden rather than graduated exposure at 16. Cited in sections 1, 3 and 4.
9. Australian Framework for Generative Artificial Intelligence (AI) in Schools, approved by Education Ministers 5 October 2023, developed by the National AI in Schools Taskforce; implemented from Term 1 2024; the 2024 review endorsed by Education Ministers in June 2025 — six principles (teaching and learning; human and social wellbeing; transparency; fairness; accountability; privacy, security and safety) and 25 guiding statements, with $1 million committed to Education Services Australia for privacy and security principles, and the position that schools should not use products that sell student data and that high-stakes marking remain human. Cited in sections 6 and 7.
10. State and territory education department AI policies and rollouts to mid-2026 — South Australia’s EdChat deployment to all public high schools; Queensland’s Corella platform rolling out statewide; Western Australia’s workload-reduction trial in eight schools; Victoria’s parental opt-in requirement for tools handling personal information and prohibition on AI-generated depictions of students and staff; Tasmania’s approved-tools-only procedure; the Northern Territory’s risk-management guidelines; and the ACT’s staged introduction. Cited in section 6.
11. Modern Movement Australia, Where Australia Stands on AI — the register of Australian AI policy, including the absence of a dedicated AI Act, the reliance on existing law, the voluntary safety standard, the paused mandatory high-risk guardrails, and the under-16 social media age ban in the online-safety chronology. Cited in sections 6 and 7.
12. Platform youth-account policies as published by Meta, TikTok, Snap and Google, 2025–26 — the application of restricted messaging, limited recommendations, content restrictions, advertising limits and stricter defaults to accounts registered as belonging to minors, and the consequence that an account registered with an adult birth date receives the unrestricted adult experience. Cited in section 3.
13. United States National Prohibition, 1920–1933 — the Eighteenth Amendment and the National Prohibition (Volstead) Act; the pattern of a sharp initial fall in consumption followed by substantial recovery; the transfer of supply to unregulated criminal production and the resulting deaths and injuries from adulterated and denatured alcohol; repeal by the Twenty-first Amendment in 1933; and the replacement regime of minimum purchase age, licensed sale, product standards, taxation and enforcement directed at sellers — the framework Australia applies to alcohol, tobacco and gambling today. Cited in section 3.
14. Prime Minister of Australia and Minister for Communications, Stronger powers and double the penalties for world-leading social media law, 28 June 2026, and contemporary reporting — the proposal to double the maximum penalty for breaches of the social media minimum age law to $99 million, stated as bringing it into line with penalties under competition and consumer law; new legislation to strengthen the eSafety Commissioner’s information-gathering powers, including compelling documents and evidence from platforms on how they enforce the ban and requiring information from third-party age-verification and digital-identity providers; the Minister’s characterisation of platform conduct as doing the bare minimum; the Prime Minister’s statement that too many children remain on social media; and the Government’s count of more than five million under-16 accounts removed, deactivated or restricted since 10 December 2025. Cited in sections 2, 5, 8 and 9.
15. eSafety Commissioner, social media age restrictions guidance, as at March 2026, and platform statements of December 2025 — the civil penalty of up to 150,000 penalty units for corporations, stated by the regulator as $54.6 million; the amendment of the legislative rules in March 2026 adding criteria for age-restricted platforms; the requirement for monthly reporting by platforms of accounts closed; and Google’s public position that the legislation was rushed and misunderstood its platform. Cited in sections 1, 2 and 5.